Pork product yield early warning system based on neural network

Abstract

The invention provides a pork product yield early warning system based on a neural network, and the system comprises a fattening pig feeding parameter collection module based on the Internet of Things, which is used for collecting the individual information of fattening pigs, adjusting the feeding amount of the fattening pigs, and detecting and controlling the multi-environment parameters of a fattening pig house by collecting the feeding parameters of the fattening pigs; the pork product yield evaluation module is used for detecting and predicting the fattening pig feeding parameters and early warning the pork product yield; wherein the pork product yield evaluation module comprises a pork product yield prediction sub-module, a fattening pig feed intake prediction sub-module, a fattening pig house temperature prediction sub-module, a fattening pig feed intake dynamic correction sub-module, a pork product yield dynamic correction sub-module and a pork product yield grade evaluator based on an ISSA-NARX neural network. According to the method, manual operation of breeding personnel is not needed, the labor cost is reduced to a great extent, and the accuracy, timeliness and comprehensiveness of prediction and early warning are improved.

Description

Technical Field

[0001] The invention relates to the technical field of neural network, and in particular to a pork product yield early warning system based on neural network. Background Art

[0002] Pork is an important food resource, and changes in its supply directly affect the market supply and demand relationship and price fluctuations. Therefore, it is necessary to continuously and timely predict and warn the output of pork products. However, in general, when predicting and warning the output of pork products, it is mostly done by farmers based on their previous farming experience. This not only has a high labor cost, but also lacks accuracy, timeliness and comprehensiveness in prediction and warning. Therefore, an intelligent solution is urgently needed. Summary of the invention

[0003] The invention analyzes the influence of fattening pig feeding parameters on fattening pork product yield, proposes a feed intake and temperature correction method for fattening pork product yield, and aims to improve the evaluation accuracy of fattening pork product yield.

[0004] The present invention provides a pork product yield early warning system based on a neural network, comprising:

[0005] The feeding parameter collection module for fattening pigs based on the Internet of Things collects the feeding parameters of fattening pigs to collect individual information of fattening pigs, adjust the feeding amount of fattening pigs, and detect and control multiple environmental parameters of fattening pig houses;

[0006] A pork product yield evaluation module, used for detecting and predicting the feeding parameters of the fattening pigs and for early warning of pork product yield;

[0007] Among them, the pork product yield evaluation module includes: a pork product yield prediction submodule, a fattening pig feed intake prediction submodule, a fattening pig house temperature prediction submodule, a fattening pig feed intake dynamic correction submodule, a pork product yield dynamic correction submodule and a pork product yield grade evaluator based on the ISSA-NARX neural network.

[0008] Preferably, the fattening pig feeding parameter collection module includes: a terminal monitoring device, a network communication layer and a data application layer;

[0009] The terminal monitoring device and the network communication layer communicate over long distances via the LoRa network;

[0010] The network communication layer and the data application layer communicate over long distances via the GPRS network.

[0011] Preferably, the terminal monitoring device is a group-raising fattening pig intelligent feeding station including: an information collection module, an information transmission module and a device control module;

[0012] The information collection module includes: temperature and humidity sensor, RFID ear tag information, trough weighing sensor, electronic floor scale;

[0013] The information transmission module includes a LoRa wireless communication module;

[0014] The equipment control module is a feed push rod motor, a wet curtain, a heater, a gas purifier, and a touch screen controlled by an STM32 single-chip microcomputer; the STM32 single-chip microcomputer is responsible for collecting and sending sensor data and controlling field equipment;

[0015] The LoRa wireless communication module sends the collected fattening pig feeding information to the LoRa aggregation node and receives the issued control instructions;

[0016] The network communication layer includes: LoRa aggregation node and GPRS-DTU module;

[0017] The LoRa aggregation node aggregates the fattening pig feeding information data collected by each LoRa terminal information collection point, and establishes a communication connection with the host computer through the GPRS-DTU module.

[0018] Preferably, the application layer includes: the pork product yield grade evaluation module, the fattening pig feeding monitoring module and the network communication interface;

[0019] The functions of the fattening pig feeding monitoring module include: pig house temperature and humidity information, fattening pig feed intake information, pork product output information, historical data statistical analysis and real-time data acquisition;

[0020] The network communication interface realizes the network communication between the upper computer monitoring platform and the lower computer measurement and control system.

[0021] Preferably, the pork product yield prediction submodule includes: a density-based pork yield cluster analysis classifier, a plurality of combined prediction models and a GRNN neural network pork product yield prediction result reconstruction model;

[0022] The pork product yields of the fattening pig feeding parameters over the years are used as inputs of the density-based pork yield cluster analysis classifier;

[0023] The density-based pork production cluster analysis classifier classifies the pork product production in previous years, and the classified pork product production value of each type is used as the input of the corresponding combined prediction model;

[0024] The combined prediction model includes: XGBoost extreme gradient boost prediction model, BiGRU bidirectional gated recurrent unit prediction model and ARIMA difference autoregressive moving average prediction model;

[0025] The prediction values ​​of the XGBoost extreme gradient enhancement prediction model, the BiGRU bidirectional gated recurrent unit prediction model, and the ARIMA differential autoregressive moving average prediction model in each group of the combined prediction models are summed up by using a genetic algorithm with variable weights to obtain a fused prediction value;

[0026] The fused prediction value of each group of the combined prediction models is used as the input of the GRNN neural network pork production prediction result reconstruction model;

[0027] The output value of the GRNN neural network pork yield prediction result reconstruction model is used as the final pork product yield output value of the pork product yield prediction submodule.

[0028] Preferably, the fattening pig feed intake prediction submodule includes: a model-based fattening pig feed intake cluster analysis classifier, multiple Elman neural network feed intake prediction models and a GRNN neural network feed intake prediction result reconstruction model;

[0029] The feed intake of fattening pigs in a period of time in the pig feeding parameters is used as an input of a fattening pig feed intake clustering analysis classifier based on a model;

[0030] The model-based fattening pig feed intake cluster analysis classifier classifies the feed intake, and the classified feed intake of each type of fattening pig is used as the input of the Elman neural network feed intake prediction model;

[0031] The feed intake prediction value of the Elman neural network feed intake prediction model is used as the input of the GRNN neural network feed intake prediction result reconstruction model;

[0032] The output value of the GRNN neural network feed intake prediction result reconstruction model is used as the final feed intake prediction output value of the fattening pig feed intake prediction submodule.

[0033] Preferably, the pig house temperature prediction submodule includes: a pig house temperature CEEMDAN decomposition model, a permutation entropy reconstruction unit, a partial autocorrelation function unit and a random forest subsequence dual-scale feature selection unit, multiple ELM neural network fattening pig house temperature prediction models and a GRNN neural network temperature prediction result reconstruction model;

[0034] The pig house temperature CEEMDAN decomposition model decomposes the pig house temperature in the feeding parameters into an intrinsic mode function IMF and a residual r(t);

[0035] The permutation entropy reconstruction unit reconstructs the eigenmode function in phase space, and uses the partial autocorrelation function unit and the random forest to perform a dual-scale feature selection unit to obtain a new subsequence, and the new subsequence is used as an input of the ELM neural network prediction model;

[0036] The output of the ELM neural network fattening pig house temperature prediction model is used as the input of the GRNN neural network fattening pig house temperature prediction result reconstruction model;

[0037] The output of the GRNN neural network fattening pig house temperature prediction result reconstruction model is used as the final fattening pig house temperature prediction output value of the fattening pig house temperature prediction submodule.

[0038] Preferably, the feed intake dynamic correction submodule for fattening pigs includes: a rate of change loop 1 and a feed intake dynamic correction SORFNN self-organizing fuzzy recursive neural network;

[0039] The output of the fattening pig feed intake prediction submodule is used as the input of the a-side of the SORFNN self-organizing fuzzy recurrent neural network for dynamic correction of feed intake;

[0040] The output of the piggery temperature prediction submodule is used as the input of the rate of change loop and the b-end input of the feed intake dynamic correction SORFNN self-organizing fuzzy recurrent neural network;

[0041] The output of the connection end of the two dy / dt of the rate of change loop 1 is used as the d-end input of the SORFNN self-organizing fuzzy recurrent neural network for dynamic correction of feed intake;

[0042] The output of the change rate loop 1 is the C-terminal input of the SORFNN self-organizing fuzzy recurrent neural network for dynamic correction of feed intake.

[0043] Preferably, the pork product yield dynamic correction submodule comprises: a rate of change loop 2 and a SORFNN self-organizing fuzzy recursive neural network;

[0044] The output of the pork product yield prediction submodule is used as the e-end input of the pork product yield dynamic correction SORFNN self-organizing fuzzy recurrent neural network;

[0045] The output of the dynamic correction submodule of feed intake of fattening pigs is used as the f-terminal input of the SORFNN self-organizing fuzzy recurrent neural network for dynamic correction of pork product yield;

[0046] The output of the connection terminal of the two dy / dt of the rate of change loop 2 is used as the h-terminal input of the SORFNN self-organizing fuzzy recurrent neural network for dynamic correction of pork product output;

[0047] The output of the change rate loop 2 is the g-end input of the SORFNN self-organizing fuzzy recurrent neural network for dynamic correction of pork product output.

[0048] Preferably, the pork product yield grade evaluator takes the size of the pork product yield prediction value output by the pork product yield dynamic correction submodule, the fattening pig breed and the pig house area as input, and outputs the pork product yield grade;

[0049] The pork product yield grades include: high pork yield, relatively high pork yield, medium pork yield, low pork yield and relatively low pork yield.

[0050] The present invention has achieved the following beneficial effects:

[0051] 1. According to the characteristics of the sample differences of pork product yield and fattening pig feed intake parameters, the present invention constructs a density-based cluster analysis classifier for pork yield and a model-based cluster analysis classifier for fattening pig feed intake, classifies the multi-point sample parameters of pork product yield and fattening pig feed intake respectively, designs multiple combined prediction models for pork product yield and a GRNN neural network pork yield prediction result reconstruction model, designs multiple Elman neural network prediction models for fattening pig feed intake and a GRNN neural network fattening pig feed intake prediction result reconstruction model; in the process of predicting pork product yield and fattening pig feed intake, their characteristics in time and space are fully considered, and data with similar causes and relatively homogeneous quality are extracted from massive data, so as to establish a more targeted prediction model that can better reflect any time period, thereby improving the prediction accuracy of pork product yield and fattening pig feed intake.

[0052] 2. The combined prediction model of the present invention establishes prediction sub-models based on three methods: XGBoost extreme gradient enhancement prediction model, BiGRU bidirectional gated recurrent unit prediction model, and ARIMA differential autoregressive moving average prediction model, which respectively realize nonlinear combination approximation of pork product output, construct a combined prediction model, and predict pork product output based on the MATLAB simulation platform. The results show that the combined prediction model can integrate the information of a single prediction method, realize information complementarity, and improve the robustness of the model. The prediction results of integrating multiple methods are more scientific and accurate than those of a single prediction method.

[0053] 3. The present invention decomposes the fattening pig house temperature parameter time series with nonlinear and non-stationary characteristics into multiple intrinsic mode function IMF components and a residual r(t) through the fattening pig house temperature CEEMDAN decomposition model. In order to avoid the significant increase in the difficulty of model training and the reduction in prediction accuracy caused by too many intrinsic mode function IMF components, the decomposed intrinsic mode function IMF components are reconstructed in phase space using permutation entropy to obtain subsequences, thereby reducing the complexity of the intrinsic mode function IMF components. The partial autocorrelation function unit and the random forest unit are used to perform double-scale feature selection on the subsequence to obtain a new subsequence, and then each new subsequence is used to construct an ELM neural network fattening pig house temperature prediction model, and finally the prediction results of each new subsequence are nonlinearly superimposed through the GRNN neural network to obtain the final fattening pig house temperature output value. Example studies show that the proposed final prediction result has a high prediction accuracy for fattening pig house temperature.

[0054] 4. The SORFNN self-organizing fuzzy recursive neural network prediction model of the present invention establishes an internal feedback connection, records the information of the previous moment and substitutes it into the next network calculation, thereby enhancing the dynamic performance of the network. The mutual information MI combined with the activation intensity is used as a self-organizing method to grow or prune regular layer neurons to enhance the generalization ability of the network, and the network parameters are adjusted using the gradient descent algorithm with an adaptive learning rate to ensure the convergence effect of the network. The example prediction results show that the model has high prediction accuracy, contains the dynamic information of the network, has strong adaptability, and has convenient adjustment capabilities using a self-organizing mechanism.

[0055] 5. The dynamic correction submodule of fattening pig feed intake of the present invention is composed of two dy / dt modules and a SORFNN self-organizing fuzzy recursive neural network. The two dy / dt units are connected in series to form a change rate loop 1. The primary change rate and secondary change rate of temperature that affect the feed intake of fattening pigs are introduced into the SORFNN self-organizing fuzzy recursive neural network training of the dynamic correction submodule of fattening pig feed intake through the change rate loop 1 to form a new input vector. It has good nonlinear mapping ability and improves the generalization ability of the network, so that it has better correction accuracy and adaptability in the dynamic correction submodule of nonlinear fattening pig feed intake than traditional static neural networks, thereby improving the prediction accuracy of fattening pig feed intake.

[0056] 6. The pork product yield dynamic correction submodule of the present invention is composed of two dy / dt modules and a SORFNN self-organizing fuzzy recursive neural network. The two dy / dt units are connected in series to form an integral loop 1. The feed intake value at the previous moment and the feed intake value at the previous two moments that affect the feed intake of fattening pigs are introduced into the SORFNN self-organizing fuzzy recursive neural network training of the fattening pig feed intake dynamic correction submodule through the integral loop 1 to generate a new input vector with good nonlinear mapping ability. The pork product yield and fattening pig feed intake input of the network model also include their feed intake at the previous moment and the previous two moments. The generalization ability of the network is improved, so that it has better prediction performance when used for time series prediction of nonlinear pork yield.

[0057] 7. The present invention dynamically corrects the prediction of pork product yield according to the feed intake of fattening pigs and the impact of the previous feed intake on pork product yield. Compared with the traditional static neural network, it has better correction accuracy and adaptive ability, and improves the prediction accuracy of pork product yield.

[0058] 8. The NARX neural network pork product yield grade evaluator based on ISSA of the present invention uses the size of the pork product yield prediction value output by the pork product yield dynamic correction submodule, the fattening pig breed and the pig house area as the input of the NARX neural network pork yield grade evaluator of ISSA. The output of the NARX neural network pork yield grade evaluator of ISSA divides the pork product yield of fattening pigs of different breeds into five grades: high pork yield, relatively high pork yield, medium pork yield, low pork yield and relatively low pork yield. The evaluator improves the accuracy and reliability of pork product yield grade evaluation.

[0059] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings.

[0060] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0062] Figure 1 This is a schematic diagram of a pork product yield early warning system according to an embodiment of the present invention;

[0063] Figure 2This is a schematic diagram of an intelligent feeding platform for fattening pigs based on the Internet of Things in an embodiment of the present invention;

[0064] Figure 3 This is a schematic diagram of a pork product yield evaluation module in an embodiment of the present invention;

[0065] Figure 4 This is a functional block diagram of the remote server monitoring software in an embodiment of the present invention;

[0066] Figure 5 It is a plan layout diagram of the LoRa terminal acquisition node and the LoRa aggregation node in an embodiment of the present invention;

[0067] Figure 6 This is a schematic diagram of an intelligent feeding station for group-rearing fattening pigs in an embodiment of the present invention. DETAILED DESCRIPTION

[0068] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0069] The present invention provides a pork product yield early warning system based on a neural network, such as Figure 1 As shown, including:

[0070] The feeding parameter collection module for fattening pigs based on the Internet of Things collects the feeding parameters of pigs to collect individual pig information and adjust the pig feeding amount, as well as detect and control multiple environmental parameters of the pig house;

[0071] A pork product yield evaluation module, used for detecting and predicting the pig feeding parameters of the fattening pigs and early warning of pork product yield;

[0072] Among them, the pork product yield evaluation module includes: a pork product yield prediction submodule, a fattening pig feed intake prediction submodule, a pig house temperature prediction submodule, a fattening pig feed intake dynamic correction submodule, a pork product yield dynamic correction submodule and a pork product yield grade evaluator based on the ISSA-NARX neural network.

[0073] The fattening pig feeding parameter collection module includes: terminal monitoring equipment, network communication layer and data application layer;

[0074] The terminal monitoring device and the network communication layer communicate over long distances via the LoRa network;

[0075] The network communication layer and the data application layer communicate over long distances via the GPRS network.

[0076] The terminal monitoring device is an intelligent feeder for fattening pigs and includes: an information collection module, an information transmission module and a device control module;

[0077] The information collection module includes: temperature and humidity sensor, RFID ear tag information, trough weighing sensor, electronic floor scale;

[0078] The information transmission module includes a LoRa wireless communication module;

[0079] The equipment control module is a feed push rod motor, a wet curtain, a heater, a gas purifier, and a touch screen controlled by an STM32 single-chip microcomputer; the STM32 single-chip microcomputer is responsible for collecting and sending sensor data and controlling field equipment;

[0080] The LoRa wireless communication module sends the collected fattening pig feeding information to the LoRa aggregation node and receives the issued control instructions;

[0081] The network communication layer includes: LoRa aggregation node and GPRS-DTU module;

[0082] The LoRa aggregation node aggregates the fattening pig feeding information data collected by each LoRa terminal information collection point, and establishes a communication connection with the host computer through the GPRS-DTU module.

[0083] The application layer includes: the pork product yield grade evaluation module, the fattening pig feeding monitoring module and the network communication interface;

[0084] The functions of the fattening pig feeding monitoring module include: pig house temperature and humidity information, fattening pig feed intake information, pork product output information, historical data statistical analysis and real-time data acquisition;

[0085] The network communication interface realizes the network communication between the upper computer monitoring platform and the lower computer measurement and control system.

[0086] The pork product yield prediction submodule includes: a density-based pork yield cluster analysis classifier, multiple combined prediction models, and a GRNN neural network prediction result reconstruction model;

[0087] The pork product yields of the fattening pig feeding parameters over the years are used as inputs of the density-based pork yield cluster analysis classifier;

[0088] The density-based pork production cluster analysis classifier classifies the pork product production in previous years, and the classified pork product production value of each type is used as the input of the corresponding combined prediction model;

[0089] The combined prediction model includes: XGBoost extreme gradient boost prediction model, BiGRU bidirectional gated recurrent unit prediction model and ARIMA difference autoregressive moving average prediction model;

[0090] The prediction values ​​of the XGBoost extreme gradient enhancement prediction model, the BiGRU bidirectional gated recurrent unit prediction model, and the ARIMA differential autoregressive moving average prediction model in each group of the combined prediction models are summed up by a genetic algorithm with variable weights to obtain a fused prediction value;

[0091] The fused prediction value of each group of the combined prediction models is used as the input of the GRNN neural network prediction result reconstruction model;

[0092] The output value of the GRNN neural network prediction result reconstruction model is used as the final pork product yield output value of the pork product yield prediction submodule.

[0093] The fattening pig feed intake prediction submodule includes: a model-based fattening pig feed intake cluster analysis classifier, multiple Elman neural network feed intake prediction models and a GRNN neural network feed intake prediction result reconstruction model;

[0094] The feed intake of fattening pigs in a period of time in the pig feeding parameters is used as an input of a fattening pig feed intake clustering analysis classifier based on a model;

[0095] The model-based fattening pig feed intake cluster analysis classifier classifies the feed intake, and the classified feed intake of each type of fattening pig is used as the input of the Elman neural network feed intake prediction model;

[0096] The feed intake prediction value of the Elman neural network feed intake prediction model is used as the input of the GRNN neural network feed intake prediction result reconstruction model;

[0097] The output value of the GRNN neural network feed intake prediction result reconstruction model is used as the final feed intake prediction output value of the fattening pig feed intake prediction submodule.

[0098] The pig house temperature prediction submodule includes: a pig house temperature CEEMDAN decomposition model, a permutation entropy reconstruction unit, a partial autocorrelation function unit and a random forest subsequence dual-scale feature selection unit, multiple ELM neural network fattening pig house temperature prediction models and a GRNN neural network temperature prediction result reconstruction model;

[0099] The pig house temperature CEEMDAN decomposition model decomposes the pig house temperature in the feeding parameters into an intrinsic mode function IMF and a residual r(t);

[0100] The permutation entropy reconstruction unit reconstructs the eigenmode function in phase space, and uses the partial autocorrelation function unit and the random forest to perform a dual-scale feature selection unit to obtain a new subsequence, and the new subsequence is used as an input of the ELM neural network prediction model;

[0101] The output of the ELM neural network fattening pig house temperature prediction model is used as the input of the GRNN neural network fattening pig house temperature prediction result reconstruction model;

[0102] The output of the GRNN neural network fattening pig house temperature prediction result reconstruction model is used as the final fattening pig house temperature prediction output value of the fattening pig house temperature prediction submodule.

[0103] The feed intake dynamic correction submodule for fattening pigs includes: a rate of change loop 1 and a feed intake dynamic correction SORFNN self-organizing fuzzy recursive neural network;

[0104] The output of the fattening pig feed intake prediction submodule is used as the input of the a-side of the SORFNN self-organizing fuzzy recurrent neural network for dynamic correction of feed intake;

[0105] The output of the piggery temperature prediction submodule is used as the input of the rate of change loop and the b-end input of the feed intake dynamic correction SORFNN self-organizing fuzzy recurrent neural network;

[0106] The output of the connection end of the two dy / dt of the rate of change loop 1 is used as the d-end input of the SORFNN self-organizing fuzzy recurrent neural network for dynamic correction of feed intake;

[0107] The output of the change rate loop 1 is the C-terminal input of the SORFNN self-organizing fuzzy recurrent neural network for dynamic correction of feed intake.

[0108] The pork product yield dynamic correction submodule includes: a rate of change loop 2 and a SORFNN self-organizing fuzzy recursive neural network;

[0109] The output of the pork product yield prediction submodule is used as the e-end input of the pork product yield dynamic correction SORFNN self-organizing fuzzy recurrent neural network;

[0110] The output of the dynamic correction submodule of feed intake of fattening pigs is used as the f-terminal input of the SORFNN self-organizing fuzzy recurrent neural network for dynamic correction of pork product yield;

[0111] The output of the connection terminal of the two dy / dt of the rate of change loop 2 is used as the h-terminal input of the SORFNN self-organizing fuzzy recurrent neural network for dynamic correction of pork product output;

[0112] The output of the change rate loop 2 is the g-end input of the SORFNN self-organizing fuzzy recurrent neural network for dynamic correction of pork product output.

[0113] The pork product yield grade evaluator takes the size of the pork product yield prediction value output by the pork product yield dynamic correction submodule, the fattening pig breed and the pig house area as input, and outputs the pork product yield grade;

[0114] The pork product yield grades include: high pork yield, relatively high pork yield, medium pork yield, low pork yield and relatively low pork yield.

[0115] The Internet of Things-based fattening pig feeding parameter collection module includes terminal monitoring equipment, network communication layer and application layer. The terminal monitoring equipment sends the detected fattening pig feeding parameters to the network communication layer through LoRa wireless communication; the LoRa aggregation node of the network communication layer aggregates the feeding parameter information sent by each terminal monitoring device, and then sends it to the upper computer server of the application layer through GPRS wireless communication; the upper computer server stores the collected fattening pig feeding parameters in the database, and the pork product production evaluation system stores the pork product production data of previous years to evaluate the pork product production grade. The entire system structure is shown in Figure 2 and Figure 3 .

[0116] The terminal monitoring equipment includes multiple LoRa terminal collection points. The LoRa terminal collection points include temperature and humidity sensors, RFID modules, feed trough weighing sensors, and electronic floor scales; the information transmission module is a LoRa wireless communication module, an STM32 single-chip microcomputer, a controlled unloading motor, a wet curtain, a heater, a gas purifier, and a touch screen. The feeding parameter information collected by the STM32 single-chip microcomputer is sent to the LoRa aggregation node through the LoRa wireless communication module. To avoid data collision, a time-division multiplexed LoRa star network is used. The software of the LoRa terminal collection point mainly realizes the collection and preprocessing of fattening pig feeding parameters, the control of pig house temperature and humidity, and LoRa wireless communication. The software is designed in C language, which is simple and intuitive, improves the readability of the program, and thus exerts the strong reliability of the equipment.

[0117] The application layer server monitoring software is based on C / S architecture and is developed in Visual Studio 2022 using C#.Net language. Its functional structure is as follows Figure 4 As shown, it includes an intelligent early warning system for pork product production, a fattening pig feeding monitoring system, and a network communication interface. The intelligent early warning system for pork product production evaluates the pork product production level based on the output of the pork product production dynamic correction submodule, the pig house area, and the fattening pig breed. The fattening pig feeding monitoring system can view and manage the historical feeding data of all fattening pigs in the pig farm. The network communication interface realizes communication between the server and the client.

[0118] In specific applications, according to the actual situation of the pig farm, the fattening pig house is networked in LoRa star shape, and the convergence gateway is arranged in the pig farm management office. Figure 5 As shown; fattening pigs are raised in group-raising fattening pig feeding stations. Each intelligent group-raising house is equipped with an intelligent feeding station. The node layout of the group-raising fattening pig intelligent feeding station is as follows Figure 6As shown, the system can realize intelligent early warning of the production of fattening pig pork products.

[0119] (1) Design of pork production prediction submodule

[0120] The pork yield prediction submodule includes: density-based pork yield cluster analysis classifier, multiple combined prediction models and GRNN neural network prediction result reconstruction model;

[0121] The historical pork production is used as input to the density-based pork production cluster analysis classifier;

[0122] The density-based pork production cluster analysis classifier classifies the pork production values ​​of the past years, and the pork product production value of each model after classification is used as the input of the corresponding combined prediction model;

[0123] A. Density-based cluster analysis classifier for pork production

[0124] The density-based pork production cluster analysis classifier used in this patent and the steps of the automatic cluster center selection method are as follows:

[0125] ①. Arrange the decision values ​​γ in descending order, with n points (i, ρ i ) is the pork product production data set S to be clustered, S1 is the first point, S n For the last point.

[0126] ②. Set the initial value of the segment point t = 2, passing through points S1 and S t and points S1, S n Draw two straight lines y1 and y2, and calculate the corresponding fitting values ​​y when x=1,2,…,n respectively * .

[0127] ③. Calculate the fitting error

[0128] ④. Let t=t+1. When t>n, jump to ⑤, otherwise jump to ②.

[0129] ⑤. Find the minimum fitting error θ t , t is the optimal number of cluster centers, and the sample points corresponding to the first t γ values ​​are the cluster centers v = [v1, v2, …, v t ].

[0130] Given a historical pork production dataset X = [x1, x2, …, x N ] T , where x i =[x i1 ,x i2 ,…,x iM], N is the number of samples, and M is the sample dimension.

[0131] The clustering algorithm steps are as follows:

[0132] ①. Obtain the natural eigenvalue sup according to the natural nearest neighbor algorithm k As the k value of KNN.

[0133] ②. Calculate the local density ρ', minimum distance δ and decision value γ of pork production samples.

[0134]

[0135] In the formula, d i,j is the sample point x i and x j The Euclidean distance between them. KNN k (x i ) is the same as x i The set of k sample points closest to each other.

[0136]

[0137] γ i =ρ i ·δ i

[0138] ③. Use the automatic cluster center selection method to select the cluster center v.

[0139] ④. Assign non-cluster center points to their corresponding clusters according to the sample allocation strategy based on the weighted KNN graph.

[0140] The density-based cluster analysis classifier for pork production divides pork production into multiple types. Each pork production type is used as the input of the corresponding combined prediction model, and the output of each combined prediction model is the fused prediction value of the pork production of that type.

[0141] B. Combined prediction model

[0142] The combined prediction model includes combined prediction model 1, combined prediction model 2 and combined prediction model N. Each combined prediction model includes XGBoost extreme gradient enhancement pork production prediction model, BiGRU bidirectional gated recurrent unit pork production prediction model and ARIMA differential autoregressive moving average pork production prediction model, and N prediction values ​​are summed up by genetic algorithm with variable weights to obtain the value of the combined prediction model; the density-based pork production clustering analysis classifier divides the pork production historical data into N types, and each type of pork production historical data is used as the input of the combined prediction model of the corresponding class. Each combined prediction model realizes the prediction of different types of pork production, thereby improving the accuracy of pork production. The output values ​​of the N combined prediction models are used as the input of the GRNN neural network pork production prediction result reconstruction model.

[0143] ①XGBoost extreme gradient enhancement pork production prediction model

[0144] XGBoost is an integrated tree model with the advantages of fast training speed and high prediction accuracy. The XGBoost model is a modeling method that predicts the future pork production of a pig farm based on the historical data of pork production of the predicted pig farm. It analyzes the time series of pork production of the predicted pig farm.

[0145] Let D = {(x i ,y i )}(|D|=n,x i ∈R m ,y i ∈R), n is the number of pork product production samples, m is the number of features contained in the sample, x i is the sample data, y i is the true value of the sample. The XGBoost model contains K CART trees, and the output is the sum of the K trees as the predicted value of the XGBoost model:

[0146]

[0147] in the formula is the predicted value of the i-th sample, i is the number of trees, F is the set of CART trees, x i represents the feature vector of the i-th data point, f k is the function model of the i-th tree.

[0148] By constructing an XGBoost model to predict the pork production of pig farms, the pork production of the pig farm of this patent can be predicted, and an XGBoost prediction model for the pork production of the corresponding pig farm can be constructed.

[0149] ②、BiGRU bidirectional gated recurrent unit pork production prediction model

[0150] Traditional recurrent neural networks (RNNs), long short-term memory (LSTMs), and gated recurrent units (GRUs) only record past information, while bidirectional gated recurrent units (BiGRUs) can retain past and future information and combine the two, allowing the prediction model to learn higher-level features from the pig farm pork production time series.

[0151] Assume that the historical data sequence of pork production in a pig farm is x = {x1, x2…, x n This patent uses BiGRU to build a pig farm pork production prediction model. BiGRU consists of two unidirectional GRUs in opposite directions. The forward GRU moves from the beginning of the sequence to the end of the sequence, and the reverse GRU moves from the end of the sequence to the beginning of the sequence. The output formula is

[0152] h t(fwd) =G fwd (x t ,h t-1(fwd) )

[0153] h t(bwd) =G bwd (x t ,h t-1(bwd) )

[0154]

[0155] In the formula, h t(fwd) is the state of the forward GRU, h t(bwd) is the state of the reverse GRU, h t(fwd) and h t(bwd) The connection symbol between t(bi) is the output of BiGRU.

[0156] By constructing a BiGRU model to predict the pork production of pig farms, the pork production of the pig farms of this patent can be predicted, and a BiGRU prediction model for the pork production of the corresponding pig farms can be constructed.

[0157] ③ARIMA difference autoregressive moving average pork production forecasting model

[0158] The ARIMA difference autoregressive moving average pork production forecasting model is a modeling method that predicts the future pork product production of a pig farm based on the historical data of pork production in the predicted pig farm. It analyzes the time series of pork product production in the predicted pig farm.

[0159] This patent uses the historical data of pig farm pork product production to analyze the time series of pig farm pork product production to study the autoregressive mean order p, the number of difference operations d, and the sliding mean order q of the time series characteristics of the ARMIA dynamic prediction model.

[0160] The ARIMA difference autoregressive moving average pork production forecasting model can be expressed as: ARMIA (p, d, q); the ARIMA dynamic forecasting equation for pig farm pork product production with p, d, q as parameters can be expressed as:

[0161]

[0162] In the formula: △ d y t for y t The sequence after d times of difference transformation, ε t is a random error with a variance constant ε t The normal distribution of i (i=1,2,…,p) and θ j (j=1,2,…,q) are the parameters to be estimated for the ARIMA difference autoregressive moving average pork production forecasting model.

[0163] ARIMA dynamic prediction of pig farm pork product output is essentially a linear prediction. The modeling steps are as follows:

[0164] Ⅰ. Time series stabilization processing: If the historical data of the pig farm's pork product output is a non-stationary time series and has a certain downward or upward trend, it is necessary to perform differential processing on the historical data of the pig farm's pork product output to become a stationary time series;

[0165] II. Model identification: The order p and q of the ARIMA dynamic prediction model for the output of pork products in pig farms are determined by using the autocorrelation function ACF and the partial autocorrelation function PACF;

[0166] Ⅲ. Model parameter estimation and testing: The maximum likelihood estimation combined with the Akaike information criterion AIC is used to obtain the best estimated values ​​of all parameters in the ARIMA dynamic prediction model for pig farm pork product production, and the significance test of the parameters, the randomness test of the residual and the white noise test are performed. Based on the test results, the feasibility of the constructed pig farm pork product production prediction model is judged. The ARIMA dynamic prediction model for pig farm pork product production with appropriate parameters is used to predict the pig farm pork product production, and the model is tested at the same time to determine whether the model is suitable. If it is not suitable, its parameters are re-estimated.

[0167] IV. Using the ARIMA model with appropriate parameters to predict the pork product output of the pig farm, this patent implements the entire modeling process of predicting the pork product output of the pig farm by calling the ARIMA module with time series analysis function in the SPSS statistical analysis software package.

[0168] C. GRNN neural network feed intake prediction model for fattening pigs

[0169] The input of the GRNN neural network fattening pig feed intake prediction result reconstruction model is the output of multiple Elman neural network fattening pig feed intake prediction models, and the output of the GRNN neural network fattening pig feed intake prediction result reconstruction model is the fusion value of the outputs of multiple Elman neural network fattening pig feed intake prediction models.

[0170] The GRNN neural network fattening pig feed intake prediction result reconstruction model constructed by this patent is a multi-input single-output network topology structure. The GRNN network consists of an input layer, a pattern layer, a summation layer and an output layer. The input vector X of the GRNN network is an n-dimensional vector, and the output vector Y of the GRNN network is a k-dimensional vector X={x1,x2,…,x n} T and Y={y1,y2,…,y k} T The number of neurons constructed in the pattern layer is equal to the number of training samples m. Each neuron corresponds to a training sample one by one. The transfer function p of the neurons in the pattern layer is i for:

[0171] p i =exp{-[(xx i ) T (xx i )] / (2*σ)},(i=1,2,…,m)

[0172] The neuron output in the formula enters the summation layer for summation. The summation layer functions are divided into two categories:

[0173]

[0174]

[0175] y in the formula i j is the jth element value in the output vector of the i-th training sample. According to the aforementioned GRNN neural network algorithm, the estimated value of the jth element of the network output vector Y is:

[0176] y j =S Nj / S D ,(j=1,2,…,k)

[0177] (2) Design of feed intake prediction submodule for fattening pigs

[0178] A. Model-based cluster analysis classifier for feed intake of fattening pigs

[0179] The model-based cluster analysis classifier for feed intake of fattening pigs is a fuzzy self-organizing neural network classifier.

[0180] This patent first uses the LVQ neural network to learn the distribution of cluster centers for fattening pig feed intake samples, and then calculates the sample density p through the number of clusters of samples and cluster centers. j The learning algorithm of the fuzzy self-organizing neural network cluster analysis classifier is as follows:

[0181] ①. Set the initial weight to an arbitrary decimal, E>0, and the number of permutation rings t=0.

[0182] ②. Input new sample x of fattening pig feed intake k =(x1,x2,…,x N ), calculate the input x k Connect weight vector w to all output nodes j The Euclidean distance of:

[0183]

[0184] ③. According to the sample density p of fattening pig feed intake j Determine the membership degree μ j :

[0185]

[0186]

[0187] In the formula, p j is the density of samples near the center point, n j is the number of times the cluster center of the jth class is selected as the nearest point.

[0188] ④. Adaptively change the learning factor α j :

[0189]

[0190] ⑤. Adjust the connection weights of all output nodes:

[0191] w ij (t+1)=w ij (t)+α j (t)×μ pj (t)×[x pi -w ij (t)]

[0192] In the formula, α j (t) is the learning factor of the jth class at time t, μ pj (t) is x p The degree of membership to the jth class.

[0193] ⑥Calculation If E tIf <E, then stop; otherwise, t = t + 1 and jump to ②.

[0194] The model - based clustering analysis classifier for the feed intake of fattening pigs classifies the historical data of the feed intake of fattening pigs into multiple types. Each type of feed intake of fattening pigs is used as the input of multiple Elman neural network prediction models, and the output of each Elman neural network prediction model is used as the predicted output value of the feed intake of this type of fattening pigs.

[0195] B. Multiple Elman neural network prediction models for the feed intake of fattening pigs

[0196] Multiple Elman neural network prediction models for the feed intake of fattening pigs input the original data of the feed intake of multiple types of fattening pigs in the pig farm to predict the future feed intake values of the original data of the feed intake of each type of fattening pigs. Each Elman neural network prediction model for the feed intake of fattening pigs can be regarded as a forward neural network with a local memory unit and local feedback connections.

[0197] The Elman neural network is also a type of dynamic recurrent neural network. Based on the BP network, it adds an associative layer in the hidden layer as a one - step delay operator to achieve the effect of memorizing the previous state, so that the system has the ability to adapt to time - varying characteristics and increases the global stability of the network.

[0198] The structure of the Elman neural network is similar to that of a 4 - layer BP neural network. The output of the hidden layer is self - connected to the input of the hidden layer through the delay and storage of the associative layer. This self - connection method makes it sensitive to the data of historical states, and the addition of the internal feedback network enhances the ability of the network to process dynamic information, thus achieving the purpose of dynamic modeling.

[0199] To effectively solve the approximation accuracy problem in the prediction of the feed intake of fattening pigs in the pig farm and enhance the role of the associative layer, this patent designs an improved Elman neural network prediction model for the feed intake of fattening pigs in the pig farm.

[0200] Let the number of nodes in the input layer, hidden layer, and output layer of the Elman neural network prediction model for the feed intake of fattening pigs in the pig farm be m, n, and l respectively. The output values of the hidden layer, associative layer, and output layer of the network are respectively:

[0201]

[0202]

[0203]

[0204] In the formula, w1 is the connection weight matrix from the associative layer to the hidden layer, w2 is the connection weight matrix from the input layer to the hidden layer, w3 is the connection weight matrix between the hidden layer and the output layer, and w4 is the connection matrix from the associative layer and the output layer to the output layer.

[0205] The number of nodes in the input layer, hidden layer and output layer of each Elman neural network fattening pig feed intake prediction model of this patent is 5, 11, and 1. The input of the Elman neural network is the classified historical data of fattening pig feed intake, and the output is the predicted value of feed intake of each type of fattening pig in the pig farm, thereby realizing the classified prediction of feed intake of fattening pigs in the pig farm and improving the prediction accuracy.

[0206] C. GRNN neural network feed intake prediction model for fattening pigs

[0207] The input of the GRNN neural network fattening pig feed intake prediction result reconstruction model is the output of multiple Elman neural network fattening pig feed intake prediction models, and the output of the GRNN neural network fattening pig feed intake prediction result reconstruction model is the fusion value of the outputs of multiple Elman neural network fattening pig feed intake prediction models. The design method of the GRNN neural network fattening pig house feed intake prediction result reconstruction model is the same as the design method of the GRNN neural network pork production prediction result reconstruction model of this patent, and will not be repeated here.

[0208] (3) Design of temperature prediction submodule for fattening pig houses

[0209] The fattening pig house temperature prediction submodule includes the pig house temperature CEEMDAN decomposition model, permutation entropy reconstruction, M ELM neural network pig house temperature prediction models and GRNN neural network pig house temperature prediction result reconstruction model. The pig house temperature CEEMDAN decomposition model decomposes the pig house temperature into the intrinsic mode function IMF and the residual r(t). The permutation entropy reconstruction unit reconstructs the decomposed intrinsic mode function in phase space, and uses the partial autocorrelation function unit and the random forest to perform a two-scale feature selection unit to obtain a new subsequence. The new subsequence is used as the input of each ELM neural network prediction model. The output of each ELM neural network pig house temperature prediction model is used as the input of the GRNN neural network pig house temperature prediction result reconstruction model. The final pig house temperature prediction value is output through the nonlinear superposition of the GRNN neural network, thereby improving the prediction accuracy of the pig house temperature.

[0210] A. CEEMDAN decomposition model of pig house temperature

[0211] During the decomposition process of the CEEMDAN algorithm, an adaptive Gaussian white noise sequence is added to the residual components, which overcomes the modal aliasing phenomenon of EMD and the Gaussian white noise problem of EEMD, and is suitable for analyzing nonlinear and non-stationary data sequences. The complete ensemble empirical mode decomposition (CEEMDAN) of adaptive white noise for fattening pig house temperature can decompose the components with different characteristics in the pig house temperature time series data information to obtain multiple IMF components and a residual r(t), reflecting the different fluctuation information of the original fattening pig house temperature signal. The adaptive Gaussian white noise added to the complete ensemble empirical mode decomposition of adaptive white noise for fattening pig house temperature increases the number of IMFs, making the decomposed components gradually smoother. The decomposition process steps of the complete ensemble empirical mode decomposition method of adaptive white noise for fattening pig house temperature time series data are as follows:

[0212] ①. Transform a Gaussian white noise w0ε i (t) is added to the original fattening pig house temperature signal x(t) to generate a new signal x i (t) = x(t) + w0ε i (t)(i∈{1,…,I})

[0213] In the formula, w0 is the noise coefficient, ε i (t) is the white noise added for the ith time, and I is the number of times the white noise is added.

[0214] ②. For a new signal x containing noise i (t) Perform EMD decomposition to obtain the first IMF component, which is the mean of the sequence components obtained after adding white noise I times.

[0215] ③. Extract the first residual component r1(t)=x(t)-IMF1(t) from the original fattening pig house temperature signal x(t).

[0216] ④. For r1(t)+w1E1(ε i (t)) is decomposed to obtain the second IMF component

[0217]

[0218] E in the formula k (·) is the kth IMF component generated after the signal is decomposed by EMD.

[0219] ⑤. Repeat steps ③ and ④, and finally the kth IMF can be obtained.

[0220] ⑥. Until the final residual signal can no longer be decomposed, the final decomposed residual signal is

[0221] In this way, the adaptive white noise complete set empirical mode decomposition model of fattening pig house temperature decomposes the pig house temperature time series data into multiple IMF components and residual r(t).

[0222] B. Permutation Entropy Reconstruction

[0223] The IMF reconstruction method based on permutation entropy is as follows:

[0224] Assume {c1(t),c2(t),…,c i (t),…,c m (t)} represents the IMFs sequence obtained after the original fattening pig house temperature signal is decomposed by CEEMDAN, then c i The phase space reconstruction vector of (t) is:

[0225] C i (t)=(c i (t),c i (t+t),…,c i (t+(n-1)t))

[0226] In the formula, t is the delay coefficient and n is the embedding dimension.

[0227] Rearrange C in ascending order according to n real numbers i (t) and satisfy:

[0228] (c i (t+(j1-1)t)≤c i (t+(j2-1)t)≤…≤c i (t+(j n -1)t))

[0229] According to (c i (t+(j1-1)t)≤c i (t+(j2-1)t)≤…≤c i (t+(j n -1)t)), C i (t) is mapped into a set of symbolic vectors:

[0230] S(ξ)=(j1,j2,…,j n )

[0231] In the formula, ξ=1,2,…,l, and l≤n!, S(ξ) is one of the arrangements in n!.

[0232] Compute the probability of a symbol vector, p1,p2,…,p l express Then c i (t) The permutation entropy value is:

[0233]

[0234] In the formula, 1 / lnn! is the regularization coefficient.

[0235] Permutation entropy can effectively measure the complexity and randomness of the temperature time series of the fattening pig house. Based on the idea of ​​similarity combination reconstruction, the original IMFs are reconstructed to generate RSCs to significantly reduce the complexity of the IMFs components. RSCs are calculated as:

[0236]

[0237] st|H pi -H p(i+j) |≤2(H pmax -H pmin ) / m

[0238] The partial autocorrelation coefficient PACF is used to calculate the possible input dimensions of all subsequences RSCs, and the random forest RF is used to remove irrelevant features, reduce the input dimension, and obtain the sequence with the optimal input dimension.

[0239] In MATLAB, the permutation entropy function PerEn(data,m,t) is called to reconstruct the IMF obtained by the CEEMDAN decomposition model of pig house temperature by permutation entropy to obtain m subsequences. Then, the partial autocorrelation function and random forest are used to perform double-scale feature selection on the subsequences to obtain M new sequences as the input of the ELM neural network fattening pig house temperature prediction model.

[0240] C. M ELM neural network temperature prediction model for fattening pig houses

[0241] The input of the ELM neural network fattening pig house temperature prediction model is the new sequence of fattening pig house temperature obtained by CEEMDAN decomposition, permutation entropy reconstruction and dual-scale feature selection. The output of M ELM neural network fattening pig house temperature prediction models is the input of the GRNN neural network fattening pig house temperature prediction result reconstruction model.

[0242] ELM extreme learning machine is a single hidden layer feedforward neural network. Compared with traditional BP neural network and support vector machine, its parameter setting is easy, convergence speed is fast, generalization ability is strong, and the effect is also very accurate. At the same time, it can overcome the problem of local minimum overfitting. ELM three-layer network structure can approximate any nonlinear function. The network model is usually divided into input layer, hidden layer and output layer. Suppose the temperature series data training sample set D = {(x i ,y i )|i=1,2,…N},x i and iare the input and output pig house temperature sample data, and N is the number of samples. Use the activation function to approximate these N samples with zero error, and the model of the ELM neural network is:

[0243]

[0244] In the formula, L is the number of hidden layer nodes, β i is the weight between the hidden layer and the output layer, w i is the weight between the input layer and the hidden layer, b i is the bias vector, g i (x) is the activation function.

[0245] D. GRNN neural network fattening pig house temperature prediction result reconstruction model

[0246] The input of the GRNN neural network pig house temperature prediction result reconstruction model is the output of multiple ELM neural network fattening pig house temperature prediction models, and the output of the GRNN neural network fattening pig house temperature prediction result reconstruction model is the fusion value of the outputs of multiple ELM neural network fattening pig house temperature prediction models. The design method of the GRNN neural network fattening pig house temperature prediction result reconstruction model is the same as the design method of the GRNN neural network pork yield prediction result reconstruction model of this patent.

[0247] (4) Design of dynamic correction submodule for feed intake of fattening pigs

[0248] The dynamic correction submodule of fattening pig feed intake is composed of two dy / dt and SORFNN self-organizing fuzzy recurrent neural network. The two dy / dt are connected in series to form the change rate loop 1. The output of the fattening pig feed intake prediction submodule is used as the input of the a terminal of the SORFNN self-organizing fuzzy recurrent neural network. The output of the fattening pig house temperature prediction submodule is used as the input of the change rate loop 1 and the b terminal of the SORFNN self-organizing fuzzy recurrent neural network. The output of the connection end of the two dy / dt of the change rate loop 1 is the input of the d terminal of the SORFNN self-organizing fuzzy recurrent neural network, and the output of the change rate loop 1 is the input of the c terminal of the SORFNN self-organizing fuzzy recurrent neural network.

[0249] The SORFNN self-organizing fuzzy recursive neural network feed intake prediction model for fattening pigs has six layers, and the algorithm of each layer is as follows:

[0250] Input layer: This layer has a total of n neurons, which pass the input values ​​directly to the next layer. Each node represents an input variable.

[0251]

[0252] Membership function layer: Gaussian membership function is used, and each node represents a membership function.

[0253]

[0254] In the formula, c ij , σ ij are the center and width of the Gaussian membership function, m is the number of fuzzy subsets, is the membership function of the mth fuzzy subset corresponding to the input variable.

[0255] Rule layer: Using multiplication fuzzy logic operators, each node represents a fuzzy logic rule.

[0256]

[0257] Recursive layer: Establish a self-feedback connection channel to enhance the dynamic ability of the network, and use a linear sum function as the activation function.

[0258]

[0259]

[0260] In the formula, j is an internal variable, and k is the current moment.

[0261] Posterity layer: execute the multiplication of TS-type fuzzy operator and the corresponding recursive variable.

[0262]

[0263] Output layer: Perform row summation on the input variables and implement defuzzification.

[0264]

[0265] In the formula, p ij are the fuzzy system parameters.

[0266] This patent constructs a TS-type fuzzy neural network FNN, establishes a feedback channel to increase the dynamic performance of the fuzzy neural network, adopts a self-organizing method based on mutual information MI and activation intensity to improve the generalization ability of the network, and uses a gradient descent method based on an adaptive learning rate to adjust network parameters to ensure the convergence effect of the network. The experimental results show that the SORFNN self-organizing fuzzy recursive neural network fattening pig feed intake prediction model has high prediction accuracy, contains dynamic information of the network, has strong adaptive ability, and uses a self-organizing mechanism with convenient adjustment capabilities.

[0267] (5) Design of dynamic correction submodule for pork production

[0268] The dynamic correction submodule of pork production is composed of 2 dy / dt and SORFNN self-organizing fuzzy recurrent neural network, and the 2 dy / dt are connected in series to form the change rate loop 2. The output of the dynamic correction submodule of fattening pig feed intake is used as the f-end input of the SORFNN self-organizing fuzzy recurrent neural network. The output of the dynamic correction submodule of fattening pig feed intake is used as the input of the change rate loop 2 and the g-end input of the SORFNN self-organizing fuzzy recurrent neural network, the output of the connection end of the 2 dy / dt of the change rate loop 2 is the input of the h-end of the SORFNN self-organizing fuzzy recurrent neural network, and the output of the change rate loop 2 is the input of the g-end of the SORFNN self-organizing fuzzy recurrent neural network. The design method of the SORFNN self-organizing fuzzy recurrent neural network pork production prediction model is the same as the design method of the SORFNN self-organizing fuzzy recurrent neural network feed intake prediction model of this patent, which will not be repeated.

[0269] (6) Design of pork yield grade evaluator based on ISSA using NARX neural network

[0270] The NARX neural network structure includes an input layer, a delay layer, a hidden layer, and an output layer. Compared with ordinary neural networks, the NARX neural network adds a feedback process from output to input, and also adds input time delay and output to input time delay, fully considering the impact of historical data on the current state, and has better adaptability to the prediction of time series data.

[0271] The output h of the jth hidden layer node j for:

[0272]

[0273] In the formula, f(·) is a nonlinear function, m and n are the delay orders of input and output respectively, and w ji is the connection weight between the i-th input and the j-th hidden layer neuron, w jl is the connection weight between the lth output and the jth hidden layer neuron, θ j is the threshold of the jth hidden layer element.

[0274] x(t) is the input of the NARX neural network prediction model, and the output y(t) of the NARX neural network prediction model is:

[0275] y(t)=f[y(t-1),y(t-2),…,y(tn),x(t-1),x(t-2),…,x(t-m+1);W]

[0276] In the formula, y(t-1), y(t-2), …, y(tn) are the model outputs after time delay, and x(t), x(t-1), …, x(t-m+1) are the model inputs after time delay.

[0277] The ISSA-based NARX neural network pork yield grade evaluator takes the output value of the pork yield dynamic correction submodule, the pig house area, and the pig breed as the input of the ISSA-based NARX neural network pork yield grade evaluator. The output of the ISSA-based NARX neural network pork yield grade evaluator is four pork yield grades: high pork yield, relatively high pork yield, medium pork yield, and low pork yield.

[0278] Pig breeds can be quantified into numbers, for example, Duroc is 1, Yorkshire is 2, Landrace is 3, Large White is 4, Pietrain is 5, etc. The pig breeds are quantified into digital quantities, and the pig house area unit is square meters. The output sizes of the ISSA-based NARX neural network pork yield grade evaluator corresponding to high pork yield, relatively high pork yield, medium pork yield and low pork yield are [1,0.8), [0.8,0.6), [0.6,0.4) and [0.4,0) respectively.

[0279] Example of design of feeding parameter collection system for fattening pigs based on the Internet of Things:

[0280] According to the actual situation of the pig farm, the fattening pig house is networked in LoRa star shape, and the convergence gateway is arranged in the pig farm management office. Figure 5 As shown in the figure, the sub-nodes of the group-raising and fattening pig intelligent feeding station are arranged in each intelligent group-raising house, such as Figure 6 As shown, this system can realize intelligent early warning of pork production of group-raised fattening pigs.

[0281] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A pork product production early warning system based on neural network, characterized in that: include: The feeding parameter collection module for fattening pigs based on the Internet of Things collects the feeding parameters of fattening pigs to collect individual information of fattening pigs and adjust the feeding amount of pigs, as well as detect and control multiple environmental parameters of fattening pig houses; A pork product yield evaluation module, used for detecting and predicting the feeding parameters of the fattening pigs and for early warning of pork product yield; Among them, the pork product yield evaluation module includes: a pork product yield prediction submodule, a fattening pig feed intake prediction submodule, a fattening pig house temperature prediction submodule, a fattening pig feed intake dynamic correction submodule, a pork product yield dynamic correction submodule and a pork product yield grade evaluator based on the ISSA-NARX neural network.

2. The pork product yield early warning system based on neural network as claimed in claim 1, characterized in that: The fattening pig feeding parameter collection module includes: terminal monitoring equipment, network communication layer and data application layer; The terminal monitoring device and the network communication layer communicate over long distances via the LoRa network; The network communication layer and the data application layer communicate over long distances via the GPRS network.

3. The pork product yield early warning system based on neural network as claimed in claim 2, characterized in that: The terminal monitoring device is an intelligent feeder for fattening pigs and includes: an information collection module, an information transmission module and a device control module; The information collection module includes: temperature and humidity sensor, RFID ear tag information, trough weighing sensor, electronic floor scale; The information transmission module includes a LoRa wireless communication module; The equipment control module is a feed push rod motor, a wet curtain, a heater, a gas purifier, and a touch screen controlled by an STM32 single-chip microcomputer; the STM32 single-chip microcomputer is responsible for collecting and sending sensor data and controlling field equipment; The LoRa wireless communication module sends the collected fattening pig feeding information to the LoRa aggregation node and receives the issued control instructions; The network communication layer includes: LoRa aggregation node and GPRS-DTU module; The LoRa aggregation node aggregates the fattening pig feeding information data collected by each LoRa terminal information collection point, and establishes a communication connection with the host computer through the GPRS-DTU module.

4. The pork product yield early warning system based on neural network as claimed in claim 2, characterized in that: The application layer includes: the pork product yield grade evaluation module, the fattening pig feeding monitoring module and the network communication interface; The functions of the fattening pig feeding monitoring module include: pig house temperature and humidity information, fattening pig feed intake information, pork product output information, historical data statistical analysis and real-time data acquisition; The network communication interface realizes the network communication between the upper computer monitoring platform and the lower computer measurement and control system.

5. The pork product yield early warning system based on neural network as claimed in claim 1, characterized in that: The pork product yield prediction submodule includes: a density-based pork yield cluster analysis classifier, multiple combined prediction models, and a GRNN neural network prediction result reconstruction model; The pork product yields of the fattening pig feeding parameters over the years are used as inputs of the density-based pork yield cluster analysis classifier; The density-based pork production cluster analysis classifier classifies the pork product production in previous years, and the classified pork product production value of each type is used as the input of the corresponding combined prediction model; The combined prediction model includes: XGBoost extreme gradient boost prediction model, BiGRU bidirectional gated recurrent unit prediction model and ARIMA difference autoregressive moving average prediction model; The prediction values ​​of the XGBoost extreme gradient enhancement prediction model, the BiGRU bidirectional gated recurrent unit prediction model, and the ARIMA differential autoregressive moving average prediction model in each group of the combined prediction models are summed up by a genetic algorithm with variable weights to obtain a fused prediction value; The fused prediction value of each group of the combined prediction models is used as the input of the GRNN neural network prediction result reconstruction model; The output value of the GRNN neural network prediction result reconstruction model is used as the final pork product yield output value of the pork product yield prediction submodule.

6. The pork product yield early warning system based on neural network as claimed in claim 1, characterized in that: The fattening pig feed intake prediction submodule includes: a model-based fattening pig feed intake cluster analysis classifier, multiple Elman neural network feed intake prediction models and a GRNN neural network feed intake prediction result reconstruction model; The feed intake of fattening pigs in a period of time in the pig feeding parameters is used as an input of a fattening pig feed intake clustering analysis classifier based on a model; The model-based fattening pig feed intake cluster analysis classifier classifies the feed intake, and the classified feed intake of each type of fattening pig is used as the input of the Elman neural network feed intake prediction model; The feed intake prediction value of the Elman neural network feed intake prediction model is used as the input of the GRNN neural network feed intake prediction result reconstruction model; The output value of the GRNN neural network feed intake prediction result reconstruction model is used as the final feed intake prediction output value of the fattening pig feed intake prediction submodule.

7. The pork product yield early warning system based on neural network as claimed in claim 1, characterized in that: The pig house temperature prediction submodule includes: a pig house temperature CEEMDAN decomposition model, a permutation entropy reconstruction unit, a partial autocorrelation function unit and a random forest subsequence dual-scale feature selection unit, multiple ELM neural network fattening pig house temperature prediction models and a GRNN neural network temperature prediction result reconstruction model; The pig house temperature CEEMDAN decomposition model decomposes the pig house temperature in the feeding parameters into an intrinsic mode function IMF and a residual r(t); The permutation entropy reconstruction unit reconstructs the eigenmode function in phase space, and uses the partial autocorrelation function unit and the random forest to perform a dual-scale feature selection unit to obtain a new subsequence, and the new subsequence is used as an input of the ELM neural network prediction model; The output of the ELM neural network fattening pig house temperature prediction model is used as the input of the GRNN neural network fattening pig house temperature prediction result reconstruction model; The output of the GRNN neural network fattening pig house temperature prediction result reconstruction model is used as the final fattening pig house temperature prediction output value of the fattening pig house temperature prediction submodule.

8. The neural network-based pork product yield early warning system according to claim 1, characterized in that: The feed intake dynamic correction submodule for fattening pigs includes: a rate of change loop 1 and a feed intake dynamic correction SORFNN self-organizing fuzzy recursive neural network; The output of the fattening pig feed intake prediction submodule is used as the input of the a-side of the SORFNN self-organizing fuzzy recurrent neural network for dynamic correction of feed intake; The output of the piggery temperature prediction submodule is used as the input of the rate of change loop and the b-end input of the feed intake dynamic correction SORFNN self-organizing fuzzy recurrent neural network; The output of the connection end of the two dy / dt of the rate of change loop 1 is used as the d-end input of the SORFNN self-organizing fuzzy recurrent neural network for dynamic correction of feed intake; The output of the change rate loop 1 is the C-terminal input of the SORFNN self-organizing fuzzy recurrent neural network for dynamic correction of feed intake.

9. The pork product yield early warning system based on neural network as claimed in claim 1, characterized in that: The pork product yield dynamic correction submodule includes: a rate of change loop 2 and a SORFNN self-organizing fuzzy recursive neural network; The output of the pork product yield prediction submodule is used as the e-end input of the pork product yield dynamic correction SORFNN self-organizing fuzzy recurrent neural network; The output of the dynamic correction submodule of feed intake of fattening pigs is used as the f-terminal input of the SORFNN self-organizing fuzzy recurrent neural network for dynamic correction of pork product yield; The output of the connection terminal of the two dy / dt of the rate of change loop 2 is used as the h-terminal input of the SORFNN self-organizing fuzzy recurrent neural network for dynamic correction of pork product output; The output of the change rate loop 2 is the g-end input of the SORFNN self-organizing fuzzy recurrent neural network for dynamic correction of pork product output.

10. The pork product yield early warning system based on neural network as claimed in claim 1, characterized in that: The pork product yield grade evaluator takes the size of the pork product yield prediction value output by the pork product yield dynamic correction submodule, the fattening pig breed and the pig house area as input, and outputs the pork product yield grade; The pork product yield grades include: high pork yield, relatively high pork yield, medium pork yield, low pork yield and relatively low pork yield.

Images